I am trying to test a mediation effect from a study that has nested data (>30 classrooms). I am doing a half-longitudinal mediation (two measurement time points). I would use the bootstrapped CI's to evaluate the indirect effects, but bootstrap does not run on nested data. I also have auxiliary variables, that help me deal with missing data and those are also not compatable with boostrapping.
Could you suggest any other method to test if indirect effect is significant? I know I cannot rely on simple p value, since it assumes that the product of a path and b path is normally distributed.
Is there a way to do Monte Carlo simulation? or perhaps some other approach?